Dimensions overall score 5.0

GitHub Code Pulse

No public code linked for this paper yet.

• Cross-attention in latent diffusion models exhibits a consistent coarse-to-fine spectral progression across prompts and seeds.result 90%
• Attention Frequency Modulation (AFM) is a training-free, plug-and-play inference-time intervention that edits cross-attention logits in the Fourier domain.method 95%
• AFM reweights low- and high-frequency bands with a progress-aligned schedule before the token softmax.technical 90%
• AFM reliably redistributes attention spectra and produces substantial visual edits while largely preserving semantic alignment.result 85%
• Token-allocation entropy mainly acts as an adaptive gain on the same frequency-based edit rather than an independent control axis.result 85%
• AFM provides a continuous handle to bias the spatial scale of token-competition patterns without retraining, prompt editing, or parameter updates.technical 90%
• Hard binary masks for frequency separation can introduce mild spatial ringing due to sharp spectral boundaries.limitation 80%
• AFM is applied only to encoder cross-attention modules in the standard configuration.method 85%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card